Study finds pro-Beijing bias in AI chatbot responses

A recent study published in the journal Nature indicates that major AI chatbots, including ChatGPT and Claude, provide significantly more favorable responses regarding Chinese leadership when queried in Chinese compared to English. Researchers suggest this discrepancy arises because AI models frequently incorporate content from Chinese state media, which dominates the available information in the Chinese language.

Experts warn that this trend reflects a broader strategy by Beijing to influence global AI development. By exporting large, state-curated data sets—such as the WanJuan collection—China aims to integrate its official narratives into the foundational data used by international developers. Analysts note that this practice allows authoritarian perspectives to shape the values and outputs of AI systems, effectively obscuring the origin of the information provided to users.

While China’s National Data Administration has announced plans to become a leading global data exporter by 2028, Western analysts express concern over the potential for these data sets to dictate the political framing of AI responses. Researchers emphasize that users may be unaware that the information they receive is derived from state-controlled sources.